Illegal Content: How We Built an AI-powered Tool to Detect It Online

For the past few months, Oxylabs team has worked on a challenge raised by the Communications Regulatory Authority of the Republic of Lithuania (RRT). RRT is a national institution that regulates electronic communications, postal, and rail markets under the European Union directives and the laws of the Republic of Lithuania.

We have already written about GovTech Lab’s challenge, and how Oxylabs won the contest to create a tool that automatically detects illegal content on the internet. In this article, we will explain step-by step, how our team of software developers, data scientists, and engineers created a powerful AI-driven tool that will make the internet cleaner. 

Before we dive into the solution, we would like to answer one question that will help understand the urgency of this sort of tool.

Why do we want to automatically detect illegal content online?

Last year, the RRT hotline in Lithuania received thousands of notifications, reporting prohibited content on the internet. About a third of them were confirmed as illegal content. In 2018, INHOPE, an International Association of Internet Hotlines, received notifications for over 220, 000 images with prohibited content. 

At Oxylabs, we are all about innovations and automation. Checking every notification by hand is inefficient. As soon as we heard about the challenge, we knew that we had the tools and the knowledge to help RRT. Our team got together to brainstorm how to find illegal content online, and we soon had a plan.

Ten weeks later, we are sharing how we built an AI-powered tool to detect illegal content online.

Solution requirements

The main requirements for the solution were:

• The solution should operate in the Lithuanian IP address range;​

• The tool must be able to identify websites operating in Lithuania;​

• It has to be able to recognize prohibited visual material;​

• The solution has to send a link (URL) of the detected prohibited visual material to the RRT hotline.

To understand the challenge better, you can find its initial description provided by the GovTech Lab.

Building the solution for detecting illegal web content

One of the main challenges was the limited time our team had to build the solution. From the beginning of the challenge, until the demo day, we had ten weeks. After the demo day, the solution will be improved, but this is the time slot we had to build a working tool for illegal content detection.

For clarity, we will explain step-by-step how the tool works.

Step one: domain and IP address check

One of the main requirements for the solution was that it has to operate in the Lithuanian IP address range. So in the first step, the tool checks Lithuanian domains and Lithuanian IP addresses.

The tool checks Lithuanian IP range, and if they have websites. The tool also checks Lithuanian domains if they resolve into Lithuanian IP addresses.

Once the list is gathered, it is time for step two: scraping content.